Analytics & Exception

---

The TruGuard Analytics page shows how governance policies and guardrails are performing in production. It is a widget-based dashboard that helps teams monitor how many requests are evaluated, how often policies are enforced, how long enforcement takes, and which policies, agents, and targets are triggering the most enforcement events. Use this page for:

• Runtime oversight – See what is being enforced right now and how much.
• Performance tuning – Find slow enforcement checks and overly broad policies.
• Security and compliance auditing – Review which actions were taken and which policies caused them.

---

​

Dashboard Widgets

The dashboard is made up of the following widgets. Each widget can be customized (see Customize the Board).

---

Requests Evaluated This widget shows the overall scale of governance activity for the selected time range.

• Requests Evaluated – The total number of requests TruGuard processed.
• Requests with Policy Enforcement – The subset of evaluated requests where at least one policy was applied.
• Policies – The number of policies being evaluated.

Compare the two request counts to understand how much traffic is actually governed. A low ratio can mean that policies are narrow, or that many requests fall outside their scope.

---

Enforcement Latency Distribution This histogram shows how long policy checks take to complete.

• X-axis – Latency buckets (for example, 0ms, 10ms, 50ms, 100ms, 200ms).
• Y-axis – Frequency, the number of checks that fell into each bucket.

Use this widget to check whether governance adds noticeable delay. If most checks fall in the lower buckets, enforcement overhead is low. A long tail toward the higher buckets means some checks are slow, and you should review the policies and validators running on those requests.

---

Enforcement Trend This time-series chart shows enforcement events over time, split by action. Filters:

• All Actions – Every enforcement event
• Block Actions – Requests blocked by a deny or block action
• Allow Actions – Requests explicitly permitted by a policy
• Human Approval – Requests that were held for an approver

Use the trend to spot sudden changes, such as a spike in blocks after a policy change or a rise in approval requests that may create a bottleneck.

---

Most Triggered Policies This ranked list shows the policies that generate the most enforcement events. For each policy, you can see:

• Policy name
• Event count – The number of enforcement events the policy caused
• Share (%) – The policy’s portion of total enforcement events

Select See More to view the full list. Policies at the top of the list are good candidates for review. Confirm that each one is intended, scoped correctly, and not blocking legitimate traffic.

---

Most Frequently Governed Agents This ranked list shows the agents that are subject to the most policy enforcement. For each agent, you can see:

• Agent name
• Event count – The number of enforcement events for that agent

Select See More to view the full list. Use this widget to find agents that need closer attention, or agents whose policies may need to be split or refined.

---

Enforcement Events by Target This widget compares enforcement across the four target types:

• Tools
• Agents
• Models
• Application

Each target shows its total number of enforcement events. Use this view to see whether enforcement is concentrated in tool calls, model calls, or application-level validators.

---

Enforcement Events by Action This widget breaks down enforcement outcomes by action, shown as counts and percentages:

• Allow
• Deny
• Mask
• Encrypt
• Other configured actions

Use this widget to understand the mix of outcomes. A high share of Deny may indicate overly strict policies. A high share of Allow may indicate policies that are not doing much enforcement.

---

Customize the Board You can tailor the dashboard to the widgets your team needs.

1. Select Customise on the board.
2. Add widgets from the pre-defined TRU GOVERN widget library.
3. If you cannot find the widget you need, ask Argus to create one. Describe the metric or view you want in plain language.

---

Validation Failures The Validation Failures view provides a request-level record of guardrail violations. Use it to investigate a specific event after you have identified a pattern in the dashboard. Each row represents a single validation failure and includes:

• Validator ID
• Date
• Category
• Stage (Input or Output)
• Prompt / Output (truncated)
• Action Taken
• Latency

The table can be searched and filtered.

---

​

Validation Failure Details

Select any failure to see the full context:

• Original Prompt / Response – The full input or output that triggered the validation.
• Exception Details
  • Validator Type – For example, Regex, PII, or Content Safety
  • Failure Reason – A human-readable explanation of why the validation failed
  • Condition – The rule or pattern that triggered the failure (for example, [A-z])
• On-Fail Action – The action configured for this validator, such as Filter, Fix, Encrypt, Allow but log, Stop execution, or Re-ask.
• Updated Prompt / Response – The content after the on-fail action was applied, if the action modified it.